convez.ai
AgenteAutomaçãoAvançado
douatwork

Gideon

douatwork/gideon

Integração cross-agent com sincronização bidirecional de conhecimento

Ver no GitHub
InicialPluginScore · de 0 a 100

Harness leve que integra assistentes de código (Claude Code, Pi Agent) como assistentes pessoais. Oferece uma skill única que cuida de scaffolding, bootstrap e sincronização bidirecional de workflows entre agentes e repositório central.

Instalação

$npx convez-ai douatwork/gideon

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Orquestrar múltiplos agentes de código com um ponto de entrada unificado
  • Sincronizar workflows e skills entre agentes automaticamente
  • Scaffolding e bootstrap de novos ambientes de runtime de agentes

Veredito

Solução bem arquitetada para operadores que precisam gerenciar múltiplos agentes de IA. O design idempotente e o foco em entrada de usuário descoberta aumentam a usabilidade, mas requer familiaridade com estrutura de agentes e git.

README

A tradução para o português deste README ainda está na fila.

Gideon — single cross-agent skill

This folder is one skill that does everything operational in Gideon from a single invocation:

  • Scaffolds the core/ stack/ taste/ workflows/ folders (plus AGENTS.md, .sync/, and a git repo) when Gideon doesn't exist yet.
  • Bootstraps a new agent runtime — writes the user-level root instruction file that points at Gideon and registers a per-agent state file.
  • Syncs bidirectionally — contributes what the agent knows back into Gideon, then derives native skills from Gideon's workflow files and records the synced commit.

It collapses Gideon's three otherwise-separate concerns (workflows/agents-init.md, workflows/agent-sync.md, and the previously-manual folder setup) into one idempotent, mode-detecting skill. Run it any time; it does only what's needed.

Why a single skill

Gideon's repo keeps agents-init.md and agent-sync.md as system instructions that each runtime reads ad hoc and never turns into a skill. That works, but it leaves the operator-facing entry points scattered and requires the user to know which document to point an agent at. Folding them into one installable skill gives every runtime the same first-class verb — invoke gideon and it figures out whether to scaffold, bootstrap, or sync. The cost is one derived artifact per runtime; the payoff is a uniform, discoverable entry point.

Installing it per runtime

SKILL.md is written runtime-agnostically (name + description frontmatter, instruction-only body). Drop it into each runtime's skills directory:

RuntimeWhere it goesNotes
Claude Code~/.claude/skills/gideon/SKILL.mdFrontmatter name/description is enough.
piruntime skills dir, gideon/SKILL.mdSame SKILL.md frontmatter contract.
Codexgideon/SKILL.md + gideon/agents/openai.yamlCodex needs the extra YAML; mirror name/description into it.
otherthe runtime's skills/commands dirAny runtime that loads a markdown skill with a name + description trigger works as-is.

Bootstrapping is a chicken-and-egg only once: copy SKILL.md into the runtime, then invoke it (e.g. "run the gideon skill"). From there it writes the root instruction file, so later sessions discover Gideon on their own and you just say "sync with Gideon" when things change.

Relationship to the canonical workflow files

This skill is a faithful superset of public/workflows/agents-init.md and public/workflows/agent-sync.md — same git-safety discipline, same bidirectional phase order, same state-file mechanics — plus the scaffold step those files assume has already happened. If the upstream sync/init contract changes, update SKILL.md to match; it is the derived artifact and Gideon's workflow files remain the source of truth.

Ver o original no GitHub

Bom para

Relacionados